[dpdk-dev] [PATCH v3 2/7] ethdev: add a link status text representation

[Ferruh Yigit]

On 6/18/2020 11:08 AM, Ivan Dyukov wrote:
> Hi Ferruh,
> 
> Thank you for the comments.
> 
> My answers are inlined.
> 
> Best regards,
> Ivan
> 17.06.2020 19:45, Ferruh Yigit пишет:
>> On 6/15/2020 10:01 AM, Ivan Dyukov wrote:
>>> This commit add function which treat link status structure
>>> and format it to text representation.
>> If I am following correctly, the initial need was to escape from speed checks
>> everytime loging link information caused by this new 'unknown' speed.
>>
>> And later suggestion was to have a pre-formatted text for link logging.
> Correct.
>>
>> This patch brings additional link status printing/formatting capability with
>> custom format string support and with new format specifiers for link (like, '%D'
>> link duplex state),
>> although this is nice work and thanks for it, I am not sure this complexity and
>> two new APIs are really needed.
> Yep.
>> For me only 'rte_eth_link_format()' without custom format support looks good
>> enough but I won't object if the concensus is to have them.
>> I am aware there are multiple applications you are updating logging slightly
>> different which requires this flexibility but what happens if they use same
>> pre-formatted text, is that difference really required or happened by time based
>> on developers taste?
> I have changed only few applications but I have plan to change all dpdk 
> examples. Even in those few apps, we have various link status logging 
> text. e.g  app/test-pmd/config.c
> @@ -600,10 +600,9 @@ port_infos_display(portid_t port_id)
>          } else
>                  printf("\nmemory allocation on the socket: 
> %u",port->socket_id);
> 
> -       printf("\nLink status: %s\n", (link.link_status) ? ("up") : 
> ("down"));
> -       printf("Link speed: %u Mbps\n", (unsigned) link.link_speed);
> -       printf("Link duplex: %s\n", (link.link_duplex == 
> ETH_LINK_FULL_DUPLEX) ?
> -              ("full-duplex") : ("half-duplex"));
> +       rte_eth_link_printf("\nLink status: %S\n"
> +                           "Link speed: %M Mbps\n"
> +                           "Link duplex: %D\n", &link);
> the status is logged in 3 lines. this is special text layoting and I 
> don't know how it will look in one line. Myabe it will require reformat 
> all screen. I don't want to change screen layoting in this patchset.

cc'ed Morten & Stephen.

To keep existing layout in the applications yes you need more flexibility, a
pre-formatted text won't cut it, but I guess I prefer your approach in v2 for
simplicity.

And I would prefer extending the one in v2 with a larger string for the
pre-formatted text, so both flexibility and the standard output can be provided,
Morten didn't like it but if I understand correctly his comment was to keep more
simple solution in ethdev and applications can do it themselves if they want
more custom log formatting, but this patch adds more complex and flexible
logging support to the ethdev.

> 
>> I will put some comments below in any case.
>>
>>> Signed-off-by: Ivan Dyukov <i.dyukov at samsung.com>
>> <...>
>>
>>> @@ -249,6 +249,9 @@ SRCS-$(CONFIG_RTE_LIBRTE_SECURITY) += test_security.c
>>>   
>>>   SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += test_ipsec.c test_ipsec_perf.c
>>>   SRCS-$(CONFIG_RTE_LIBRTE_IPSEC) += test_ipsec_sad.c
>>> +
>>> +SRCS-$(CONFIG_RTE_LIBRTE_ETHER) += test_ethdev_link.c
>> +1 to unit test.
>>
>> <...>
>>
>>> +int
>>> +rte_eth_link_printf(const char *const fmt,
>>> +		    struct rte_eth_link *link)
>>> +{
>>> +	char text[200];
>>> +	int ret;
>>> +	ret = rte_eth_link_format(text, 200, fmt, link);
>> Will it be paranoid to add "text[199] = 0" to be sure any custom 'fmt' won't
>> cause any harm?
> 
> rte_eth_link_format already do that and it must do that. I would prefer to don't hide rte_eth_link_format bugs in rte_eth_link_printf function.

OK, makes sense.

What do you think putting an assert (RTE_ASSERT) to check that formatted buffer
is proper C string?

> 
>>
>>> +	printf("%s", text);
>> Not sure if the error still should be printed on error case?
>> Like for example what the code does when fmt="%X"?
> yep. I'll add 'ret' check.
>>
>>> +	return ret;
>>> +}
>>> +
>>> +int
>>> +rte_eth_link_format(char *str, int32_t len, const char *const fmt,
>>> +		    struct rte_eth_link *link)
>> Why not have the 'len' type 'size_t'?
> 
> Yes, I can change type of the len but internally we have 'int32_t clen = 
> len;'
> defined below.  clen should be signed variable because rte_eth_link_format
> much more simpler then snprintf. e.g. snprintf may be called with
> snprintf(buff, 0, "some text %d", val); no errors returned in this case.
> it returns length of formated string.so internally I use signed clen
> to detect end of line and avoid uint overflow.
> rte_eth_link_format with incorrect or short buffer returns error.

I see, using 'strlcat' and 'strlcpy' may help here, since both does get full
size of the buffer as parameter which removes the need of offset calculations
and possible negative values there. And both returns positive values, again
prevents using signed variables to capture return value.
So I think this can make possible to only use size_t variable as length.

> 
>>
>>> +	int offset = 0;
>>> +	int32_t clen = len;
>>> +	const char *fmt_cur = fmt;
>>> +	double gbits = (double)link->link_speed / 1000.;
>>> +	/* TBD: make it international? */
>>> +	static const char LINK_DOWN_STR[]     = "Link down";
>>> +	static const char LINK_UP_STR[]       = "Link up at ";
>>> +	static const char UNKNOWN_SPEED_STR[] = "Unknown speed";
>>> +	static const char MBITS_STR[]	      = "Mbit/s";
>>> +	static const char GBITS_STR[]	      = "Gbit/s";
>>> +	static const char AUTONEG_STR[]       = "Autoneg";
>>> +	static const char FIXED_STR[]         = "Fixed";
>>> +	static const char FDX_STR[]           = "FDX";
>>> +	static const char HDX_STR[]           = "HDX";
>>> +	static const char UNKNOWN_STR[]       = "Unknown";
>>> +	static const char UP_STR[]            = "Up";
>>> +	static const char DOWN_STR[]          = "Down";
>>> +	if (str == NULL || len == 0)
>>> +		return -1;
>>> +	/* default format string, if no fmt is specified */
>>> +	if (fmt == NULL) {
>>> +		if (link->link_status == ETH_LINK_DOWN)
>>> +			return snprintf(str, (size_t)clen, "%s", LINK_DOWN_STR);
>>> +
>>> +		offset = snprintf(str, (size_t)clen, "%s", LINK_UP_STR);
>>> +		if (offset < 0 || (clen - offset) <= 0)
>>> +			return -1;
>>> +		clen -= offset;
>>> +		str += offset;
>>> +		if (link->link_speed == ETH_SPEED_NUM_UNKNOWN) {
>>> +			offset = snprintf(str, clen, "%s",
>>> +					  UNKNOWN_SPEED_STR);
>>> +			if (offset < 0 || (clen - offset) <= 0)
>>> +				return -1;
>> better to use 'strlcpy' & 'strlcat', they are easier to use for these kind of
>> checks.
> OK
>>
>> <...>
>>
>>> +	/* Formated status */
>>> +	} else {
>>> +		char c = *fmt_cur;
>>> +		while (c) {
>>> +			if (clen <= 0)
>>> +				return -1;
>>> +			if (c == '%') {
>>> +				c = *++fmt_cur;
>>> +				switch (c) {
>>> +				/* Speed in Mbits/s */
>>> +				case 'M':
>>> +					if (link->link_speed ==
>>> +					    ETH_SPEED_NUM_UNKNOWN)
>>> +						offset = snprintf(str,
>>> +						  clen, "%s",
>>> +						  UNKNOWN_STR);
>>> +					else
>>> +						offset = snprintf(str,
>>> +						  clen, "%u",
>>> +						  link->link_speed);
>> Code readiblity is not great here because you hit the 80char limit, this is a
>> sign that something is wrong like function is already too long.
>> Can you please try to fix this, like extracting some part of the code to its own
>> function or return after end of the 'if' statement which can save one more level
>> indentation etc...
> agree. I'll define one static function and move part of the code to it. 
> It should reduce indent.
>>
>> <...>
>>
>>> diff --git a/lib/librte_ethdev/rte_ethdev.h b/lib/librte_ethdev/rte_ethdev.h
>>> index 2090af501..83291e656 100644
>>> --- a/lib/librte_ethdev/rte_ethdev.h
>>> +++ b/lib/librte_ethdev/rte_ethdev.h
>>> @@ -2295,6 +2295,58 @@ int rte_eth_link_get(uint16_t port_id, struct rte_eth_link *link);
>>>    */
>>>   int rte_eth_link_get_nowait(uint16_t port_id, struct rte_eth_link *link);
>>>   
>>> +
>>> +/**
>>> + * print formated link status to stdout. This function threats all
>>> + * special values like ETH_SPEED_NUM_UNKNOWN, ETH_LINK_DOWN etc. and convert
>>> + * them to textual representation.
>>> + *
>>> + * @param fmt
>>> + *   Format string which allow to format link status. If NULL is provided
>>> + *   , default formating will be applied.
>>> + *   Following specifiers are available:
>>> + *    - '%M' link speed in Mbits/s
>>> + *    - '%G' link speed in Gbits/s
>>> + *    - '%S' link status. e.g. Up or Down
>>> + *    - '%A' link autonegotiation state
>>> + *    - '%D' link duplex state
>>> + * @param link
>>> + *   Link status provided by rte_eth_link_get function
>>> + * @return
>>> + *   - Number of bytes written to stdout. In case of error, -1 is returned.
>> Does it worth to mention the log still will be printed on error?
> I'll change the function. It will print nothing on error.
>>
>>> + *
>>> + */
>>> +int rte_eth_link_printf(const char *const fmt,
>>> +			struct rte_eth_link *link);
>>> +
>>> +/**
>>> + * Format link status to textual representation. This function threats all
>>> + * special values like ETH_SPEED_NUM_UNKNOWN, ETH_LINK_DOWN etc. and convert
>>> + * them to textual representation.
>>> + *
>>> + * @param str
>>> + *   A pointer to a string to be filled with textual representation of
>>> + *   device status.
>>> + * @param len
>>> + *   Length of available memory at 'str' string.
>>> + * @param fmt
>>> + *   Format string which allow to format link status. If NULL is provided
>>> + *   , default formating will be applied.
>>> + *   Following specifiers are available:
>>> + *    - '%M' link speed in Mbits/s
>>> + *    - '%G' link speed in Gbits/s
>>> + *    - '%S' link status. e.g. Up or Down
>>> + *    - '%A' link autonegotiation state
>>> + *    - '%D' link duplex state
>>> + * @param link
>>> + *   Link status provided by rte_eth_link_get function
>>> + * @return
>>> + *   - Number of bytes written to str array. In case of error, -1 is returned.
>>> + *
>>> + */
>>> +int rte_eth_link_format(char *str, int32_t len, const char *const fmt,
>>> +			struct rte_eth_link *eth_link);
>>> +
>> These new APIs needs to be experimental by process (__rte_experimental).
>>
>> Need the add these APIs to the .map file (rte_ethdev_version.map), so that they
>> will be exported in the dynamic library (.so).
>>
>>
>